Multiple Convolutional Neural Networks for Online Visual Tracking
Tianli Sun, Cairong Zhao, Kang Chen · 2018
The paper presents a new online visual tracking method based on convolutional neural network (CNN). Our algorithm consists of multiple CNN tracker units focusing on different object features, and each unit is trained with a proper number of domains determined by our method. For the model update process, traditional CNN online re-training is avoided, replaced by a timely tracker switching strategy. At each moment, only one main tracker predicts and others stand by to ensure efficiency. The experimental results show that our method can track objects in most scenarios, and implicate that our CNN model has a generalization ability in distinguishing target and background.